Application of Hybrid Data Mining Model (Genetic Algorithm-Wavelet-Deep Neural Network-Monte Carlo Method) for Forecasting the Price of Agricultural Products: A Case Study of Future Price of Saffron in Agricultural Commodity Exchange

The price forecasting and its changes trend is one of the most important factors in decision making and formulating strategies related to agricultural products. This study aimed at presenting a hybrid data mining model for accurate price forecasting of agricultural products, including nonlinear models of wavelet transform, genetic algorithm, deep neural network and Monte Carlo technique. This proposed model involved a two-stage hybrid model and the base model of nonlinear-nonlinear. In this proposed model, the genetic algorithm for determining the optimal lag of price time series, the wavelet function for the de-noising of price data, the deep neural network for price forecasting, the Monte Carlo method for simulating the most probable price probability and finally, the complex soft calculations for "out-of-sample forecasting with new data set" were used. Results of comparison of the proposed model including "Genetic Algorithm-Wavelet Transform-Deep Neural Network-Monte Carlo", through evaluation criteria, with three competing models of "Genetic Algorithm-Deep Neural Network-Monte Carlo", "Genetic Algorithm-Wavelet Transform-Neural Network-Monte Carlo" and "Genetic Algorithm-Neural Network-Monte Carlo" showed that the proposed model had the better performance in forecasting of future price of saffron compared to the three competing models. Also, the use of deep neural network compared to neural network, the application of wavelet theory for de-noising and also the use of Monte Carlo technique to simulate the predicted prices, increase the forecasting accuracy of future price of saffron. In addition, the use of soft calculations showed that the proposed model had the necessary efficiency and high accuracy for short-term forecasting of the future price of saffron. Therefore, the present study has a good position in achieving the index of maximum accuracy, scenario making of future price trends, sensitivity analysis of components affecting the price and finally, forecasting the future price. Accordingly, the use of the proposed model to forecast the price of agricultural products is recommended.